AI Orchestration: Building Persistent, Context-Aware Workflows

The promise of AI lies not just in its ability to suggest, but to do. The conversation with JJ Englert on "How I AI" reveals that the true power of tools like Claude Cowork isn't in replacing human thought, but in orchestrating it with unprecedented efficiency. The non-obvious implication is that by embracing AI as a project manager and executor, knowledge workers can transcend the limitations of individual tasks and build persistent, context-aware systems that automate complex workflows. This isn't about using AI for isolated queries; it's about constructing a personal operating system where AI acts as a distributed team, handling the grunt work and providing layered feedback. Those who adopt this project-based, agent-orchestrating approach will gain a significant advantage by offloading cognitive load and accelerating their output, transforming their day-to-day work from a series of discrete actions into a cohesive, AI-augmented workflow.

The Project as a Persistent AI Partner

The shift from simple chat interfaces to project-based AI interaction represents a fundamental change in how we can leverage artificial intelligence for daily work. Instead of isolated conversations, Claude Cowork allows users to define persistent "projects" -- essentially folders on their computer -- that serve as the context for ongoing AI tasks. This approach is crucial because it imbues the AI with lasting memory and understanding, moving beyond the ephemeral nature of single chat threads.

JJ Englert highlights that a project in Cowork is fundamentally a folder, a concept familiar to anyone in business. This democratization of AI interaction means that individuals, regardless of their technical background, can become "AI orchestrators." By organizing work into these project folders, users can build a "brain file" -- a detailed document outlining preferences, team members, and working styles. This file acts as a foundational prompt, ensuring that the AI, Claude, has a consistent understanding of the user's needs and context from the outset. This eliminates the repetitive need to re-establish context, a common frustration with traditional AI chat interfaces.

"This way, once I trigger a prompt within Co-Work, it gets all of this information with it and just knows so much more about me. Sometimes we had, in ChatGPT, we had that one thread of like, 'Oh, this is my thread for all my product marketing,' and, 'This is my thread for brainstorming about this,' or whatever. You had that one thread with a lot of good context that was very important to you, but you couldn't transfer it anywhere. If you set up the time with your folders, you'll be able to build those threads, but in a very transferable way that allows you to take that context and bring it with you every single time you want to do anything in Co-Work. That is the big unlock here with Co-Work. You're not starting over from the start."

This persistent context is what enables AI to move from a reactive assistant to a proactive partner. When a project is established, all subsequent tasks within that project share the same memory. This means Claude can recall previous interactions, understand the evolving needs of the project, and provide more coherent and relevant outputs over time. This is a stark contrast to single-session chats where context is lost, forcing users to re-explain their situation repeatedly. The ability to build and transfer this context is the "big unlock," allowing for continuous progress without the drag of re-establishing foundational knowledge.

The Power of Connectors and Personalized Skills

A significant advancement in Claude Cowork is its ability to connect to external tools and data sources through "connectors." These integrations, ranging from Gmail and Slack to Notion and Google Calendar, allow the AI to not only process information but to act upon it. This is where the promise of AI doing work, rather than just suggesting it, truly materializes.

The immediate benefit is the ability to automate mundane tasks. For instance, analyzing sent emails to build a personalized writing skill is a powerful example. By feeding Claude a history of your emails, it can learn your unique tone, style, and phrasing, enabling it to draft future communications that sound authentically like you. This moves beyond generic AI-generated text to highly personalized output, saving significant time and improving the quality of communication.

"So this is going to start building an email skill for me, and it's going to do it by connecting to my Gmail, looking at my outbox of all the emails that I've sent in the last 30 days, analyzing that writing skill, and just perfectly matching the tone that I use in my writing. You know, AI is used for writing in a lot of different ways, and of course, there's a lot of AI slap, but when you can feed AI like a series of like 100 messages, it could really get close to your exact writing style, and that's what this is doing here."

The nuance here is the concept of "progressive trust." Users can start by having the AI draft emails, then progress to having it read emails, and eventually grant it more extensive permissions. This gradual increase in AI autonomy, coupled with granular control over permissions (e.g., "always ask before sending"), allows users to build confidence in the system while maximizing productivity. This layered approach to trust is critical for widespread adoption, especially among those less technically inclined. The ability to create personalized "skills" -- reusable sets of instructions tailored to specific tasks and user preferences -- further amplifies this, turning repetitive actions into automated processes.

Orchestrating AI with Sub-Agents and Scheduled Tasks

The concept of "orchestration" is central to maximizing the value of Claude Cowork. This involves not just assigning tasks to a single AI agent but managing multiple agents and coordinating their efforts. Englert introduces two key mechanisms for this: sub-agents and scheduled tasks.

The "sub-advisory board" technique, where multiple AI agents with distinct personas review a piece of work, is a powerful application of systems thinking. By simulating feedback from different stakeholders--like a boss, a customer, or an ideal customer profile (ICP)--users can gain a comprehensive understanding of their work's potential reception before it's even finalized. This layered feedback mechanism is particularly valuable for solo operators or small teams who may lack diverse perspectives. It allows for pre-emptive refinement, reducing the risk of delivering work that misses the mark.

"When you use sub-agents, it will spin up three different agents, and each of those agents can have their own persona with a fresh context window, meaning a fresh perspective, to go and look at your work in an objective way. If you're a product manager, build it and put your boss in there, your engineering partner, and your customer. Say, 'Every PRD review from these three points of view and give me feedback.'"

Scheduled tasks, such as a daily morning debrief, represent another layer of sophisticated automation. By setting up a task that runs at a specific time, the AI can proactively gather information from connected tools--email, Slack, calendar--and compile a daily plan or briefing. This transforms the AI from a tool you use into a system that works for you on a regular cadence. This transforms the morning routine from a reactive scramble to an informed start, powered by AI that has already processed key information. The persistent context of the project ensures that these scheduled tasks are not generic but are tailored to the specific goals and ongoing work within that project. This is where the true competitive advantage lies: not in being the fastest to type a prompt, but in building systems that continuously deliver value with minimal ongoing intervention.

Key Action Items

• Establish Project Folders: For any significant ongoing work, create dedicated folders on your computer to serve as project bases. This is the foundational step for persistent AI context.
• Develop a "Brain" File: Within each project folder, create an .md file that details your working preferences, key stakeholders, and desired AI interaction style. This file will serve as your primary prompt for establishing AI context.
• Connect Essential Tools: Integrate key business applications (email, calendar, messaging platforms) using Claude Cowork's connectors. Prioritize tools that handle high-volume, repetitive communication or information retrieval.
• Build Personalized Writing Skills: Leverage email and messaging history to train AI agents to write in your unique voice. This is an immediate payoff for improving communication efficiency and consistency.
• Implement Sub-Agent Review Systems: For critical outputs (e.g., PRDs, marketing copy, project plans), set up sub-agents with distinct personas (e.g., ICP, manager, peer) to provide multi-faceted feedback. This requires upfront effort but yields higher quality outputs and reduces revision cycles.
• Automate Daily Debriefs: Configure scheduled tasks to run daily, compiling information from connected tools into a morning briefing or daily action plan. This creates a proactive AI assistant that prepares you for the day ahead.
• Embrace Progressive Trust: Gradually increase the AI's permissions and autonomy as you become more comfortable. Start with drafting and reading, then move towards more automated actions, always maintaining oversight. This is a longer-term investment in unlocking deeper AI capabilities.